biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[VLM] Simplify post-processing of replacement info (#12269)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-01-22 08:48:13 +08:00
committed by GitHub
parent 09ccc9c8f7
commit df76e5af26
10 changed files with 175 additions and 208 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/models/aria.py
View File

@@ -342,13 +342,7 @@ class AriaProcessingInfo(BaseProcessingInfo):
return self.get_hf_config().vision_config
def get_hf_processor(self):
processor = self.ctx.get_hf_processor(AriaProcessor)
# Patch for https://github.com/huggingface/transformers/issues/35768
processor.tokenizer.image_token = "<|img|>"
processor.image_token = "<|img|>"
return processor
return self.ctx.get_hf_processor(AriaProcessor)
def get_supported_mm_limits(self) -> Mapping[str, Optional[int]]:
return {"image": None}
@@ -381,7 +375,7 @@ class AriaDummyInputsBuilder(BaseDummyInputsBuilder[AriaProcessingInfo]):
}
hf_processor = self.info.get_hf_processor()
image_token: str = hf_processor.image_token # type: ignore
image_token: str = hf_processor.tokenizer.image_token # type: ignore
return ProcessorInputs(
prompt_text=image_token * num_images,